English 
French 
ChatGPT has become one of those tools people don’t realize they depend on until it stops responding mid-task. A blank screen, a spinning cursor that never resolves, or a vague error message, and suddenly you’re not sure if the problem is your Wi-Fi, your browser, or OpenAI’s servers on the other end.
The good news is you can usually figure out which one it is in under a minute, and what you do next depends entirely on the answer.
Step 1: Check OpenAI’s official status page
OpenAI maintains a public status page at status.openai.com that tracks the health of ChatGPT, the API, DALL-E, Whisper, and other services separately. Each component gets its own status, so the API can be fully operational while the consumer ChatGPT web app is having problems, or the other way around. If you’re specifically wondering about the API rather than the chat interface, check the component that matches what you’re actually using.
The page also logs incident history, so if something looks off, you can usually find a posted update explaining what’s affected and, once there’s an ETA, how long it’s expected to take.
One limitation worth knowing about: status pages are updated by people, not automatically the instant something breaks. During a fast-moving incident, there’s often a short lag between when users start noticing problems and when the status page reflects it. If the page shows “all systems operational” but you’re clearly having trouble, that doesn’t automatically mean the issue is on your end. It might just mean the incident hasn’t been posted yet.
Step 2: Cross-check with an independent monitor
Because official status pages can lag, it helps to check an independent source that monitors OpenAI’s services from outside their own infrastructure. Third-party uptime tools ping services from multiple locations around the world and report back based on what they actually observe, not on what a company has chosen to publish.
The same logic applies to your own connection. If you want to rule out a broader internet or DNS problem before you assume it’s OpenAI, running a quick check through a website down checker confirms whether the issue is isolated to one service or part of something wider on your network.
Step 3: Try a different device or network
If the status page says everything’s fine, but ChatGPT still won’t load, switch to your phone’s mobile data instead of Wi-Fi, or try a completely different device. If it works there, the problem is local: your router, your Wi-Fi, a VPN, or something specific to the browser or app you were using before. If you’re not sure whether the issue is isolated to ChatGPT or something broader with your connection, checking whether a site is down right now against a couple of other services you use daily is a quick way to tell the two apart.
Common ChatGPT and OpenAI error messages, explained
Not every hiccup means the whole service is down. A lot of the time, the error message itself tells you exactly what’s going on if you know how to read it.
“ChatGPT is at capacity right now.” This means demand is temporarily exceeding available server capacity, usually during peak usage hours or right after a major product announcement drives a spike in traffic. It’s not the same as a full outage. It typically resolves within minutes as capacity frees up or scales to meet demand.
A response that stops generating mid-sentence. This is often a timeout on unusually long responses rather than a service-wide problem. Regenerating the response or breaking your request into smaller parts usually works around it.
A vague “Something went wrong” message. This one is frustrating because it doesn’t say much, but it usually points to a temporary hiccup in processing your specific request rather than a broader outage. Refreshing the page and trying again resolves it most of the time.
429 Too Many Requests. This is rate limiting, not downtime. If you’re using the API, it means you’ve exceeded the number of requests allowed for your account tier in a given time window. If you’re using the consumer ChatGPT app, it usually means you’ve hit a usage cap tied to your plan. Either way, this is specific to your account and has nothing to do with whether the service is generally available.
5xx errors from the API (500, 502, 503). These indicate a server-side problem on OpenAI’s end, distinct from account-level rate limiting. If you’re building on the API, this is the category worth setting up automatic retries for, since it usually clears up on its own within a short window.
What to do while it’s down
If you’ve confirmed the outage is real and not local to you, there isn’t much to do except wait it out, but a few things make the wait less painful.
Check the status page periodically rather than repeatedly refreshing ChatGPT itself. Reposting the same request over and over during an active outage doesn’t help it recover faster, and on the API side, it can actually make rate limiting worse once service resumes, and there’s a backlog to work through.
If you’re in the middle of something time-sensitive, it’s worth having a backup plan. Keeping a second AI assistant or tool in your back pocket for genuinely urgent moments isn’t a bad habit, the same way you’d have a backup way to send an important email if your primary provider went down.
If you’re a developer relying on the API in production, this is exactly the scenario retry logic with exponential backoff exists for. Rather than hammering the API immediately after a failure, wait briefly, retry, and increase the wait time between each subsequent attempt if it keeps failing. Checking the rate limit headers returned with each response also tells you your real quota state in real time, instead of guessing.
When the problem is actually on your end
Before you assume OpenAI is down, it’s worth ruling out a handful of local causes that produce very similar symptoms.
Clear your browser’s cache and cookies, especially if the page loads but won’t respond to input. A corrupted local session can look identical to a server-side outage from the user’s side.
Disable browser extensions, particularly ad blockers, privacy tools, or script blockers, since these sometimes interfere with the way ChatGPT’s web app sends and receives data.
Check whether a VPN is active. Some VPN endpoints get temporarily blocked or rate-limited due to abuse from other users sharing that same IP, which has nothing to do with your account or OpenAI’s actual availability.
Check your account status directly. Expired billing, a paused subscription, or hitting a plan’s usage limit all produce error messages that can look like an outage but are really account-specific.
Getting ahead of it next time
If ChatGPT or the OpenAI API is central to your workflow, whether that’s a personal habit or something your business depends on, it’s worth setting up your own monitoring instead of finding out about outages the hard way. Our guide on how website monitoring actually works covers the basics if you’re new to the concept, and understanding how tools detect outages in seconds is useful context for setting realistic expectations about how fast you’ll actually know.
For a broader monitoring setup that isn’t specific to one AI provider, our free website monitoring service lets you track uptime for any service or site you rely on, so you’re not caught off guard the next time something you depend on goes offline.
Check OpenAI’s official status page first, then cross-check with an independent monitoring tool since status pages can lag behind real incidents by several minutes. If both show normal status but you still can’t connect, try a different device or network to rule out a local issue.
It means current demand is exceeding available server capacity at that moment, usually during peak hours. It’s a temporary condition, not a full outage, and it typically resolves within minutes without any action needed on your part.
Yes, they’re different services with independent infrastructure, and OpenAI’s status page tracks them as separate components. It’s entirely possible for the API to be fully operational while the ChatGPT web app has problems, or the reverse.
No. A 429 means you’ve hit a rate limit or usage cap tied to your specific account or API key, which is unrelated to whether the broader service is available. It’s the system telling you to slow down, not telling you something is broken.
It varies widely depending on the cause. Capacity-related slowdowns often clear up within minutes, while infrastructure-level incidents can take longer to fully resolve. Checking the status page for posted updates is the most reliable way to get a real estimate rather than guessing.